You are on page 1of 3

Weekly Progress Report

Date: Mar 18th, 2021


Name: Hagai Raja Sinulingga

I. This week progress


 Watch and generate PPT from Prof. Andrew Ng “Machine Learning” lecture videos
 Learn deep learning from “Deep Learning for Computer Vision with Python” book by Dr. Adrian
Rosebrock

1. Machine Learning lecture videos


*Video progress:

Total learned: 112/112 videos (+37 on this week), ppt done: 8/112 (+4 on this week), summary:

(blue = done on previous week, green = done in this week; yellow = partially done; white = not any)

Lectures Total vids PPT Can be assigned as Material


1 3 Week 1 Intro, overview of machine learning
2 8 Week 2 Linear regression one variable
3 6 Optional Linear algebra
4 7 Week 3 Linear regression multiple variable
5 7 Week 4 Octave (demo) -> could be replaced by reading docs
6 7 Week 5 Logistic regression
7 4 Homework Regularization
8 6 Week 6 Neural networks representation
9 8 Week 6 Neural networks learning
10 7 Week 7 Advice for applying machine learning
11 5 Week 7 Machine learning system design
12 6 Week 8 Support vector machines -> a lot of linear algebra
13 5 Week 9 Clustering
14 7 Week 10 Dimensionality reduction
15 8 Week 11 Anomaly detection
16 6 Week 12 Recommender system
17 6 Week 13 Large scale machine learning
18 4 - Application example: Photo OCR -> get cutted (incomplete)
19 1 - Conclusion summary
2. Learn Deep Learning for Computer Vision with Python
*What I have learnt

 Ch.10 Neural Network Fundamentals

1
In this chapter I learned the idea behind neural network (NN), feed-forward, and backpropagation
algorithm. Then I learned how to implement basic NN from scratch with python and train it to solve
bitwise problem OR, AND, and XOR. From the experiment it is proven that NN can solve non-linear
classification problem like XOR, but it has to be done by multiple layer NN. Then I also learn
experimenting the NN with Keras library to learn from MNIST (recognize number from a handwriting
picture) and CIFAR-10 (classify 10 classes of picture based on the object depicted inside). It could
get high accuracy on MNIST but not on CIFAR-10. Then the writer explained about the 4 ingredients
of NN which are: dataset, loss function, model/architecture, and optimization method. At last, he
explained about the several ways (constant, normal distribution, LeCun uniform, Glorot/Xavier
uniform, MSRA uniform) to do weight initialization.

 Ch.11 Convolutional Neural Network

Here I learned how to compute convolution from an image, implement it on python, and what the role
in Convolutional Neural Network (CNN). Then I learned type of layer in CNN which are:

o Convolutional (CONV), layer that do convolution of kernels to the image. The parameters
are depth, stride, and padding
o Activation (ACT or RELU), same like in basic activation layer in basic NN
o Pooling (POOL), layer to do similar like CONV but here it only finds maximum or average
of the block instead of multiplying and sum
o Fully Connected (FC), layer which all neurons inside are connected to all activations from
previous layer
o Batch Normalization (BN), layer to normalize the activations values before passing it to
next layer so it will stabilize batch training
o Dropout (DO), a form of regularization by randomly drop some connections from incoming
activation to avoid a neuron being too dominant

After that I also learned the common layer patterns that used to build CNN and rule of thumb when
designing new architecture. Then the chapter ended with discussion about CNN behavior to invariant
in image like translation, rotation, and scaling. The answer is as an individual filter it can’t, but as a
whole it might can, especially the translation where pooling layer could help much. The exact answer
is relative to the dataset and experimentation during model learning.

II. Next week planning


1. Continue generate ppt slides
2. Continue learning from deep learning for computer vision vol 1 ch.12-15

2
III. Discussion
 On generating ppt, focus on content first. If the animation needs a lot of effort, just keep it as note
for future consideration.

IV. References
[Machine Learning Youtube videos] https://www.youtube.com/watch?v=FCUBwP-
JTsA&list=PLLssT5z_DsK-h9vYZkQkYNWcItqhlRJLN

Rosebrock, Adrian. (2019). Deep Learning for Computer Vision with Python Starter Bundle 3rd edition.
PyImageSearch.

You might also like